-
Bug
-
Resolution: Duplicate
-
P4
-
None
-
None
-
None
The specjvm2008::crypto.signverify workload have scalability issue for the high vCPU number. According to the JFR data it has issue with the 'Lock Instances' with stack trace:
================
void sun.security.provider.NativePRNG$RandomIO.implNextBytes(byte[])
void sun.security.provider.NativePRNG.engineNextBytes(byte[])
void java.security.SecureRandom.nextBytes(byte[])
BigInteger sun.security.provider.DSA.generateK(BigInteger)
byte[] sun.security.provider.DSA.engineSign()
byte[] java.security.Signature$Delegate.engineSign()
byte[] java.security.Signature.sign()
byte[] spec.benchmarks.crypto.signverify.Main.sign(byte[], String, PrivateKey)
================
The total blocking time changed from the ~2 seconds for the 128 vCPU to ~7 hours for the 384 vCPU.
================
void sun.security.provider.NativePRNG$RandomIO.implNextBytes(byte[])
void sun.security.provider.NativePRNG.engineNextBytes(byte[])
void java.security.SecureRandom.nextBytes(byte[])
BigInteger sun.security.provider.DSA.generateK(BigInteger)
byte[] sun.security.provider.DSA.engineSign()
byte[] java.security.Signature$Delegate.engineSign()
byte[] java.security.Signature.sign()
byte[] spec.benchmarks.crypto.signverify.Main.sign(byte[], String, PrivateKey)
================
The total blocking time changed from the ~2 seconds for the 128 vCPU to ~7 hours for the 384 vCPU.
- duplicates
-
JDK-8278371 Scalability bottleneck in NativePRNG
- Open
-
JDK-8278371 Scalability bottleneck in NativePRNG
- Open